On 11 June 2026 the European Parliament approved the amendments to the AI Act contained in the digital omnibus package, by 423 votes to 57, with 174 abstentions. The abstention count is worth noting: it is large enough to read as a position rather than as absence.
The timetable, now settled
Obligations for stand-alone high-risk systems apply from 2 December 2027, and from 2 August 2028 for high-risk systems embedded as safety components in regulated products. Watermarking obligations apply from 2 December 2026 for systems placed on the market before 2 August 2026. Compliance with the ban on child sexual abuse material and non-consensual intimate imagery is required by 2 December 2026.
The Commission does some of the work for you
In June the Commission produced freely available icons for labelling AI-generated content under Article 50(4), published the Code of Practice on Transparency of AI-Generated Content, and released two sets of FAQs on implementing and signing it. Signatory submissions were due by 22 July 2026.
An official icon set sounds like a small thing. For a small provider it removes a design decision and a legal risk simultaneously: a label that regulators have published is a label no one has to argue about.
Machine-readable consent
Also in June, the actress Cate Blanchett launched a tool called the Human Consent Registry in the European Parliament. It allows an individual to declare, in machine-readable form, whether and how AI systems may use their likeness.
Whatever one makes of a celebrity launch, the underlying construct is precisely the sort of object this Lab exists to study: a legal permission expressed in a format a system can act on directly, rather than a clause a human has to read, interpret and honour. Consent that only exists as prose is consent that scales badly. Encoding it is a legal-design problem before it is a technical one, and it is the same problem that document automation runs into from the other direction.
Drawn from The EU AI Act Newsletter #105, 29 June 2026.